Discrimination of species within the Enterobacter cloacae complex using MALDI-TOF Mass Spectrometry and Fourier-Transform Infrared Spectroscopy coupled with Machine Learning tools

Enterobacter cloacae complex (ECC) encompasses heterogenic genetic clusters of species that have been associated with nosocomial outbreaks. These species may host different acquired antimicrobial susceptibility patterns and their identification is challenging. DNA-based techniques are laborious and require specific equipment. MALDI-TOF MS has showed low accuracy for the discrimination of ECC species. The aim of this study is to develop machine learning predictive models using MALDI-TOF MS and new diagnostic technologies like Fourier-Transform Infrared Spectroscopy (FTIR-S) for species-level identification of these species. A total of 163 ECC clinical isolates were included in the study: 47 for the predictive model development and internal validation and 126 for external validation. All spectra obtained by MALDI-TOF MS and FTIR-S were processed using Clover MS Data Analysis software. Two models were created: Model A for differentiate six ECC species and Model B for E. hormaechei subspecies. For MALDI-TOF MS spectra, Model A identified correctly 96.0% of isolates using Random Forest (RF) algorithm, and Model B identified 94.1% using Support Vector Machine (SVM). Regarding FTIR-S, Model A identified 73.0% of isolates by RF and Model B 72.5%. Two new predictive models were created for FTIR-S: Model C for discrimination of E. hormaechei from non-E. hormaechei (87.3% identification, RF) and Model D for differentiation among non-E. hormaechei species (62.7% identification, RF). MALDI-TOF MS combined with machine learning tools could be a rapid and accurate method for species-level identification within the ECC. FTIR-S differentiated general groups of the ECC although discrimination of non-E. hormaechei species was poor.